Exploring the "Repeating Inputs" problem

[Paul_Lamb]

@Tachion

you’re referencing the columbs paper, but at this point this is second hand knowledge that might be wrong

@Tachion

exactly the type of problem which the TBT solved

I should point out that TBT currently also relies on the “output layer” concept from the Columns paper. If you question the validity of this concept, then you must also question TBT itself (at least in the theory’s current form) From the Frameworks paper, section " Rethinking Hierarchy, the Thousand Brains Theory of Intelligence":

Every cortical column learns models of complete objects. They achieve this by combining input with a grid cell-derived location, and then integrating over movements (see Hawkins et al., 2017; Lewis et al., 2018 for details). This suggests a modified interpretation of the cortical hierarchy, where complete models of objects are learned at every hierarchical level, and every region contains multiple models of objects (Figure 6B).

The reference to “Lewis et al. 2018” is the Columns Plus paper, which goes into more technical detail on object recognition and grid cells. From this paper, it discusses the method of voting between many sensory patches:

In previous work (Hawkins et al., 2017) we showed how these networks work together by using an additional population of cells which represents the current object, invariant to the sensor’s location. Each of these networks votes on this classified object. In this way, multiple instantiations of this network work together to recognize objects using the input from multiple independent moving sensors.

This “additional population of cells” is of course is a reference to the output layer described in the Columns paper. Hopefully this demonstrates that what I have described above isn’t a major deviation from TBT, but rather it borrows a concept from it.